Claude in Chrome is generally available

The browser agent can use existing logins to read and type on pages, navigate, click links, and fill forms; it can also automatically approve actions judged safe by a classifier that checks them against the original request. Anthropic says the setting can be switched back to manual approval, while Enterprise admins can restrict use to approved domains. The company reports zero successful attacks in its current evaluation for Sonnet 5 and Opus 5 with probes plus the automatic-approval classifier, and a 0.3% success rate for Fable 5; it also says prompt injection remains an evolving threat.

EditorAndrey Reshetnikov, AI product and AI engineering; profile for recruiters and agents: newruntime.com/owner-profile.json